White House orders federal AI-testing unit CAISI to stop publishing model evaluations

Trump administration officials, including National Cyber Director Sean Cairncross, directed the Center for AI Standards and Innovation (CAISI) — the main U.S. government body that tests frontier AI models — to halt publication of its assessments pending implementation of President Trump's June 2, 2026 AI security executive order. The order shifts model evaluation from CAISI's public process toward a classified framework run by national-security agencies, after the agency had already published more than 40 model evaluations that served as a shared public baseline. Companies will still submit models for review, but results will largely remain behind closed doors.

  • Sean Cairncross (National Cyber Director)
  • Trump administration officials

Federal scientific and standards agencies serve the public, not the party in power, and the assessments they produce are meant to be public goods that researchers, lawmakers, and citizens can scrutinize. The Center for AI Standards and Innovation had built a public track record of more than forty evaluations giving everyone a shared, independent baseline for what frontier AI models can actually do. Ordering the agency to stop releasing that work and routing it into a classified process removes public agency data on political grounds and blocks the distribution of agency research. That is the kind of suppression that lets officials control what the public is permitted to know about technologies bearing directly on national safety and rights.

  1. White House reins in AI testing unit as national security concerns growWall Street Journal primary accessed June 10, 2026
  2. CAISI ordered to stop public model evaluations amid new AI executive orderCrypto Briefing secondary accessed June 10, 2026
  3. Early Edition: June 10, 2026Just Security secondary accessed June 10, 2026
  4. Center for AI Standards and Innovation (CAISI)NIST secondary accessed June 10, 2026